Cancer Classification With MicroRNA Expression Patterns Found By An Information Theory Approach

Some non-coding small RNAs, known as microRNAs (miRNAs), have been shown to play important roles in gene regulation and various biological processes. The abnormal expression of some specific miRNA genes often results in the development of cancer. In this paper, we find discriminatory miRNA patterns for cancer classification from miRNA expression profiles with an information theory approach. Our approach evaluates subset of miRNAs by checking the mutual information between these miRNAs and the class attribute I(X;Y ) with respect to the entropy of the class attribute H(Y ). Then, optimal subset of miRNAs that satisfies I(X;Y ) = H(Y ) or H(Y )− I(X;Y ) ≤ ×H(Y ) for noisy data sets are chosen to build the classification models. The experimental results show that the expression patterns from a small set of miRNAs are very accurate in prediction. Further, the experimental results also suggest that the expression patterns of these informative miRNAs are conserved in different vertebrates during the evolution process.